METHOD FOR IDENTIFYING MISLABELED TRAINING DATA WITH THE AID OF UNLABELED DATA

摘要

PURPOSE: it is a kind of to identify that accidentally labelling training data is provided to the label of estimation unlabeled data using unlabeled data, pass through estimation fail data and above-mentioned unlabeled data in training data to identify and accidentally label. ;CONSTITUTION: label about unlabeled data is estimation (S601). The training dataset that unlabelled data and the label of estimation include. Noise is detected, by filtering training data set (S603). Noise detected removes. The noise eliminates training data set output (S605). ;The 2011 of copyright KIPO submissions